LEK autumn 2026: 60 questions from outside the CEM question base. We counted all 200

For years, people have said the public CEM question base gives you “about 70% of the exam”. In August we tested that claim on two sessions and found it was a large underestimate. We then counted the paper of 5 September 2026 the same way, question by question. This time the folk number turned out to be exactly right. That is news, not confirmation.

One note for readers sitting the exam in English: the English paper is the same test. We compared all 200 questions in both languages and found identical answer keys, identical answer statistics and identical option counts. CEM translates the same items. Everything below applies to you.

Results

We compared each of the 200 questions against the public CEM base and against the 17 archived LEK sessions in our bank.

Where the question came from	Count	Share
In the CEM base and on at least one earlier exam	108	54%
In the CEM base, never on an exam before	32	16%
From the public base in total	140	70%
Never seen anywhere, completely new	60	30%

One caveat up front: the CEM base did not change between May and September. We checked the modification dates of all twenty files (part one: 20–21 May 2026). Those 140 questions were available to everyone three and a half months before the exam.

Why 70% is a surprise

Because the five previous sessions looked different. Same method, same sources:

Session	From the public CEM base	Outside the base
spring 2024	93%	14
autumn 2024	96%	8
spring 2025	96%	7
autumn 2025	93%	13
spring 2026	96%	8
autumn 2026	70%	60

The number of repeats from earlier exams did not change: 108, against 93–98 in previous sessions. What almost disappeared was the category of “base only” questions. In spring 2026 there were 96 of them; now there are 32. In their place CEM put 60 questions nobody had seen before.

Whether this is a one-off or a new rule, February 2027 will tell. For now it is worth knowing that a single session was enough to cut the predictable part of the exam by 26 percentage points.

The new questions are much harder

CEM publishes, for every question, the percentage of correct answers and a discrimination index. We set them against the origin of each question.

Origin	Count	Correct answers	Discrimination	Questions below 50% correct
From the public base	138	84%	0.29	0
New	60	56%	0.19	22

Not a single question from the base troubled the majority of candidates. Among the new questions, 22 did. The average candidate collected about 116 points from the base and about 34 from the new questions. The base alone was enough to pass with room to spare, since the threshold is 112 points. It was not enough for a good place in the ranking.

There is one more observation, a more cautious one. Six of the new questions have a discrimination index of zero or below. That means candidates with the highest overall scores did no better on them than everyone else. This is how a question usually behaves when it measures something other than knowledge: it is ambiguous, its key is debatable, or it is badly worded. We do not say which six. We show the numbers.

Where the new questions sat

This is the most interesting thing about this paper, and you cannot see it without lining up the question numbers. The new questions were not scattered across the test. They stood together, at the start of every subject block:

Numbers	Block	New questions
1–9	internal medicine	9
40–49	paediatrics	10
69–79	surgery and oncology	11
96–104	obstetrics and gynaecology	9
122–131	internal medicine, second part	10
142–143	psychiatry	2
156–160	emergency medicine	5
186–189	medical certification	4

The rest of each block came from the base. Anyone working through the paper in order met the hardest questions on entering each new section, after which it got easier. Worth remembering when managing time at the next session: a question you do not recognise at the start of a block does not mean the whole block will be like that.

By subject, the new questions clustered in two areas: internal medicine (20) and oncology (14). Together that is 57% of all new questions. In August we wrote that when CEM adds a fresh question, it most often does so in exactly those two fields. This session confirmed it emphatically.

The three questions that defeated the most candidates

The lowest percentage of correct answers in the whole paper. All three are new. We give the CEM key and the distribution of answers, quoting the official English wording.

No. 72, internal medicine. Correct: 8.6%.

Patient malnutrition is one of the major factors increasing postoperative risk. Which of the following features is a criterion for diagnosing malnutrition in a patient? A. BMI <20. B. albumin level <6 g/dL. C. reduced food intake for 2 weeks. ← CEM key D. weight loss >5% over 6 months. E. none of the above.

The most popular choices were E (37%) and B (32%). Discrimination: 0.08. This was the hardest question of the session, and one of those on which the strongest candidates did no better than anyone else.

No. 122, internal medicine. Correct: 15.3%.

Common causes of hemoptysis do not include: A. lung cancer. B. tuberculosis. C. bronchitis. D. bacterial pneumonia. E. bleeding diathesis. ← CEM key

68% of candidates chose bronchitis. A negatively worded question that, under exam conditions, gets read as “what causes haemoptysis”. The classic way to lose a point on a question you had the knowledge to answer.

No. 46, paediatrics and neurology. Correct: 18.1%.

A mother presents to the Emergency Department with her 4-year-old son weighing 20 kg, who experienced a generalized tonic-clonic seizure lasting approximately 4 minutes, which resolved spontaneously before the arrival of the emergency medical team. The child has a history of epilepsy and had similar seizures in the past. His status after the seizure is stable, with no signs of central nervous system infection and there are no indication for hospitalization. As part of the discharge recommendations, the doctor plans to prescribe a rescue medication for rectal administration to be used in the event of a prolonged epileptic seizure. What dose of diazepam in the form of a rectal enema should be recommended for this child? A. 10 mg. ← CEM key B. 7.5 mg. C. 5 mg. D. 2.5 mg. E. 1 mg.

Answers split between 5 mg (35%) and 2.5 mg (31%). Unlike the previous two, this question discriminated properly (0.14): candidates who knew the dosing got it right.

What came back

The repeats looked as they always do. 21 questions had appeared six months earlier, in the spring 2026 session, and 18 in autumn 2025. Two questions on this paper had already appeared in six earlier sessions: the doctor’s duty to notify the police or the prosecutor when certifying a death (no. 177), and the deadline for reporting a vaccine adverse event, 24 hours (no. 133). More than 85% of candidates answered both correctly.

The seven-time record holders from our earlier analysis, the Quicktrach question and the Mallampati scale, took this session off.

Two questions were cancelled

CEM cancelled questions 82 and 148. Both came from the public base and both had a key there.

Question 82 asked for the most common cause of mechanical ileus in the lower gastrointestinal tract; the key in the base pointed to peritoneal adhesions. Question 148 asked for the most frequent side effect of clozapine; the key in the base pointed to cardiac arrhythmia. In both cases the key did not survive contact with the candidates. The conclusion is simple and uncomfortable: the base is the best study source there is, but it is not infallible, and a base question with a wrong key can only be recognised by someone who understands the topic rather than remembers the letter.

What follows from this

The base still gets you a pass. 140 questions at an average 84% success rate is about 116 points against a threshold of 112. Anyone who worked through the base and nothing else got through.

The base stopped deciding the ranking. In spring 2026 there were 8 questions outside the base, and the places in the top quarter were decided on them. In autumn there were 60. That is where the ranking was decided, not on the base, and mostly in internal medicine and oncology.

Do not panic at the start of a block. New questions sit at the beginning of each section. Fewer than half the field solved 22 of them. A question you have never seen is normal and says nothing about the rest of the block.

Negatively worded questions and questions with a number cost points regardless of knowledge. On the haemoptysis question, 68% of candidates chose the answer that would have been correct without the negation. Those losses can be trained away, unlike 60 new topics.

How we counted

Matching works on the question text together with all its answer options: a question with the same stem but changed options counts as new. In addition, we checked each of the 60 new questions by fuzzy matching against our entire bank of 4,183 LEK questions. We found four reworkings of known stems with different answer options. They stayed in the “new” category.

For sessions before May 2026 we compare against a version of the base that postdates the exam, so their “from the base” share may be slightly inflated by questions added after the fact. For autumn 2026 this problem does not arise. The direction of the change is certain; the exact scale for older sessions is not.

CEM had not yet published the result statistics for the autumn 2026 session (mean score, pass rate, results by university) on the day of publication. We will cover them separately.

The autumn 2026 session is now in the app

All 200 questions from 5 September 2026, with the CEM key and answer statistics, are available in kursnalek alongside the other 17 sessions, in Polish and in the official English translation. Questions from the base come with explanations; the new questions are receiving them progressively.

Try the free LEK exam in English →

---

Data sources: the LEK paper of the autumn 2026 session with its answer key and psychometric data, published by the Centre for Medical Examinations, CEM (cem.edu.pl); the public LEK question base (cem.edu.pl, as of 20–21 May 2026); archived LEK papers from 17 sessions (2013–2015, 2021–2026) compiled in the kursnalek bank. Percentage of correct answers and discrimination index (point-biserial correlation) according to CEM.